AgentSkillsCN

browser-skill

使用真实浏览器浏览网页，直观查看页面内容，并与页面元素进行交互。

SKILL.md
--- frontmatter
name: browser-skill
description: Use a real browser to navigate the web, see pages visually, and interact with elements.

Overview

The Browser Skill allows you to control a real, headless Chrome browser instance. This is useful for:

  • •Navigating to websites.
  • •"Seeing" the page structure via semantic snapshots.
  • •Interacting with elements (click, type, fill, scroll).
  • •Handling dynamic content (JS) that standard POST/GET requests miss.

Workflow

  1. •Open: open <url> to navigate.
  2. •Inspect: snapshot -i to get a list of interactive elements with ID references (e.g., @e1).
  3. •Interact: Use the ID references to interact.
    • •click @e1
    • •fill @e1 "search query"
    • •press @e1 Enter

Commands

CommandArgumentsDescription
open<url>Navigate to a URL.
back-Go back in history.
forward-Go forward in history.
reload-Refresh the page.
snapshot['-i']Get a semantic list of interactive elements (@e refs).
click<ref>|<selector>Click an element (e.g., @e1).
dblclick<ref>|<selector>Double-click an element.
fill<ref>|<selector> <text>Clear and fill an input field.
type<ref>|<selector> <text>Type text naturally (useful for auto-complete).
press<ref>|<selector> <key>Press a specific key (e.g., Enter, ArrowDown).
hover<ref>|<selector>Hover over an element.
scroll[dx, dy]Scroll the viewport (default: down).
click_coords<x> <y>Click at specific coordinates.

Tips

  • •Always use snapshot -i before interacting to get fresh @e references.
  • •@e references are ephemeral and may change after navigation or dynamic updates.
  • •Use fill for standard forms, type for complex inputs reacting to keystrokes.